Aniso smg #172

pdziekan · 2024-02-20T11:02:42Z

uses formulas from limpdata++ branch aniso_smg

… direction

pdziekan · 2024-09-24T13:01:09Z

TODO:

aniso smg as run/compiletime option
fix tke calculation (wrong now, because mix_len has changed)

… issues

Copilot

Pull request overview

This PR refactors the SGS (SubGrid-Scale) model implementation to support multiple schemes using formulas from the limpdata++ branch aniso_smg. It replaces the boolean sgs flag with a string parameter that accepts "iles", "smg", or "smgani", enabling selection between Implicit LES, isotropic Smagorinsky, and anisotropic Smagorinsky models.

Changes:

Refactored SGS model selection from boolean to string-based scheme selection
Implemented anisotropic Smagorinsky model alongside existing isotropic version
Updated test infrastructure to support multiple SGS schemes with separate output directories
Modified mixing length calculations to support both isotropic and anisotropic approaches

Reviewed changes

Copilot reviewed 28 out of 45512 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
tests/unit/api_test.cpp	Added test name parameter for unique output directories per test
tests/unit/CMakeLists.txt	Refactored test definitions using helper function and added smgani test
src/uwlcm.cpp	Changed sgs parameter from boolean to string, updated validation logic
src/solvers/slvr_sgs.hpp	Refactored into base class with isotropic/anisotropic variants using template specialization
src/solvers/slvr_common.hpp	Updated profile recording for new mixing length variables
src/solvers/blk_1m/slvr_blk_1m_common.hpp	Simplified inheritance to use refactored slvr_sgs
src/run_hlpr_*.cpp	Updated function signatures to accept string sgs parameter
src/run_hlpr.hpp	Updated function declaration for string sgs parameter
src/run_hlpr.cpp	Added handling for smg/smgani/iles schemes and updated set_profs call
src/opts/opts_lgrngn.hpp	Updated reference from mix_len to sgs_eddy_size
src/detail/profiles.hpp	Added new mixing length profile arrays for iso/hori/vert schemes
src/cases/*.hpp	Updated set_profs signatures to accept nps array instead of nz
src/cases/CasesCommon.hpp	Implemented mixing length calculations for isotropic and anisotropic models
.github/workflows/test_uwlcm_hlpr.yml	Updated CI to test all three SGS schemes separately

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

src/solvers/slvr_sgs.hpp

tests/unit/CMakeLists.txt

Copilot · 2026-01-19T13:43:11Z

src/cases/CasesCommon.hpp

+        // size of SGS eddies, used in SD SGS models (note that they are isotropic!)
+        profs.sgs_eddy_size = min(max(k, 0.5) * dz * 0.845, sgs_delta);
+
+        // isotropic mixing length - used in isotropic SMG and in SD SGS models
+        // like in Smagorinsky 1963, but with default characteristic length scale (sgs_delta) equal to dz (not dx*dy*dz)^(1/3))
+        // similar approach used e.g. in the SAM model
+        // also there's a reduction near ground 
+        // NOTE: it's not multiplied by the smagorinsky constant here (that is done in slvr_sgs), because mix_len_iso is used also in SD SGS (opts_lgrngn.hpp) where no such constant is used
+        // profs.mix_len_iso = min(max(k, 0.5) * dz * 0.845, sgs_delta);
+        // profs.mix_len_iso_sq = pow2(smg_c * min(max(k, 0.5) * dz * 0.845, sgs_delta));


The magic number 0.845 appears multiple times (lines 162, 171, 180, 181) without explanation. While there's a comment on line 167 mentioning reduction near ground, it would be helpful to document what this constant represents (appears to be related to von Kármán constant or mixing length reduction factor).

Suggested change

// size of SGS eddies, used in SD SGS models (note that they are isotropic!)

profs.sgs_eddy_size = min(max(k, 0.5) * dz * 0.845, sgs_delta);

// isotropic mixing length - used in isotropic SMG and in SD SGS models

// like in Smagorinsky 1963, but with default characteristic length scale (sgs_delta) equal to dz (not dx*dy*dz)^(1/3))

// similar approach used e.g. in the SAM model

// also there's a reduction near ground

// NOTE: it's not multiplied by the smagorinsky constant here (that is done in slvr_sgs), because mix_len_iso is used also in SD SGS (opts_lgrngn.hpp) where no such constant is used

// profs.mix_len_iso = min(max(k, 0.5) * dz * 0.845, sgs_delta);

// profs.mix_len_iso_sq = pow2(smg_c * min(max(k, 0.5) * dz * 0.845, sgs_delta));

// empirical factor reducing SGS eddy size / mixing length near the ground

const real_t sgs_near_ground_reduction = real_t(0.845);

// size of SGS eddies, used in SD SGS models (note that they are isotropic!)

profs.sgs_eddy_size = min(max(k, 0.5) * dz * sgs_near_ground_reduction, sgs_delta);

// isotropic mixing length - used in isotropic SMG and in SD SGS models

// like in Smagorinsky 1963, but with default characteristic length scale (sgs_delta) equal to dz (not dx*dy*dz)^(1/3))

// similar approach used e.g. in the SAM model

// also there's a reduction near ground (see sgs_near_ground_reduction above)

// NOTE: it's not multiplied by the smagorinsky constant here (that is done in slvr_sgs), because mix_len_iso is used also in SD SGS (opts_lgrngn.hpp) where no such constant is used

// profs.mix_len_iso = min(max(k, 0.5) * dz * sgs_near_ground_reduction, sgs_delta);

// profs.mix_len_iso_sq = pow2(smg_c * min(max(k, 0.5) * dz * sgs_near_ground_reduction, sgs_delta));

Copilot · 2026-01-19T13:43:12Z

src/solvers/slvr_sgs.hpp

  protected:

-  real_t prandtl_num;
+  real_t prandtl_num, karman_c; // karman_c used only in aniso...


The inline comment is incomplete: '...used only in aniso...' should be expanded to clarify that karman_c is only used in the anisotropic Smagorinsky model.

Suggested change

real_t prandtl_num, karman_c; // karman_c used only in aniso...

real_t prandtl_num, karman_c; // karman_c is used only in the anisotropic Smagorinsky model

Copilot · 2026-01-19T13:43:13Z

src/cases/CasesCommon.hpp

+        profs.mix_len_hori_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dxy)));
+        profs.mix_len_vert_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dz)));


The mixing length formula is complex and spans one line. Consider breaking it into intermediate variables or adding a comment explaining the formula structure (appears to be harmonic mean of two mixing lengths).

Suggested change

profs.mix_len_hori_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dxy)));

profs.mix_len_vert_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dz)));

// combine wall-based and grid-based length scales using a harmonic mean of their squared values

real_t wall_mix_len = karman_c * max(k, 1) * dz;

real_t hori_grid_mix_len = smg_c * dxy;

real_t vert_grid_mix_len = smg_c * dz;

real_t inv_wall_len_sq = real_t(1) / pow2(wall_mix_len);

real_t inv_hori_grid_len_sq = real_t(1) / pow2(hori_grid_mix_len);

real_t inv_vert_grid_len_sq = real_t(1) / pow2(vert_grid_mix_len);

real_t hori_mix_len = real_t(1) / (inv_wall_len_sq + inv_hori_grid_len_sq);

real_t vert_mix_len = real_t(1) / (inv_wall_len_sq + inv_vert_grid_len_sq);

profs.mix_len_hori_sq = pow2(hori_mix_len);

profs.mix_len_vert_sq = pow2(vert_mix_len);

Copilot · 2026-01-19T13:43:14Z

src/cases/CasesCommon.hpp

+        // real_t(1) / (real_t(1) / pow(this->smg_c * params.di, 2) + (*this->params.mix_len)(this->vert_idx))
+        // profs.mix_len_vert_sq = dz; // real_t(1) / pow2(karman_c * max(k, 1) * dz);
+        // add dx dy terms; also add sgs_delta, from user params...
+        // TODO: save these, use them in slvr_sgs, ...


These commented-out lines and TODOs should be removed or implemented. They appear to be development notes that should not remain in production code.

Suggested change

// TODO: save these, use them in slvr_sgs, ...

Copilot · 2026-01-19T13:43:15Z

src/solvers/slvr_sgs.hpp

+    this->k_m[0](this->ijk) = where(this->k_m[0](this->ijk) > 0.1 * pow(this->params.di, 2) / this->dt, 0.1 * pow(this->params.di, 2) / this->dt, this->k_m[0](this->ijk));
+    this->k_m[0](this->ijk) = where(this->k_m[0](this->ijk) < 1e-6 * pow(this->params.di, 2),  1e-6 * pow(this->params.di, 2), this->k_m[0](this->ijk));


The magic numbers 0.1 and 1e-6 for clamping k_m lack explanation. A comment indicating these are numerical stability bounds (as mentioned in 'min/max as in Simon and Chow 2021') would be helpful.

pdziekan and others added 4 commits February 14, 2024 17:06

anisotropic SMG with mixing lengths dependant on resolution along the…

be2d05d

… direction

k_ma limiters

58d5af3

Merge branch 'blk_trapez_fixes' into aniso_smg

dc6a24a

Merge branch 'master' into aniso_smg

d3d61c0

pdziekan and others added 12 commits November 21, 2024 11:12

Merge branch 'master' into aniso_smg

dc6d008

central gradient in subsidence instead of forward to fix top boundary…

f092d8b

… issues

fix boost phoenix linkage for boost 1.81-1.83

b47512f

Merge branch 'master' into aniso_smg

5a7e8cb

comment

98b32c0

Merge branch 'master' of github.com:igfuw/UWLCM into aniso_smg

d8b0144

run_hlpr: sgs as string option accepting iles,smg,smagni

37a5e0d

smganiso WIP

65a2b48

slvr_sgs: smg_common class

d0f5882

Merge branch 'master' of github.com:igfuw/UWLCM into aniso_smg

7f7c425

smgani solver

1bb554f

smgani unit test

488abcc

pdziekan marked this pull request as ready for review January 19, 2026 13:40

Copilot AI review requested due to automatic review settings January 19, 2026 13:40

pdziekan changed the title ~~[WIP] Aniso smg~~ Aniso smg Jan 19, 2026

Copilot AI reviewed Jan 19, 2026

View reviewed changes

GA test smgani fixes

8e8b257

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Aniso smg #172

Aniso smg #172

Uh oh!

pdziekan commented Feb 20, 2024

Uh oh!

pdziekan commented Sep 24, 2024

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Copilot AI Jan 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

	real_t prandtl_num, karman_c; // karman_c used only in aniso...
	real_t prandtl_num, karman_c; // karman_c is used only in the anisotropic Smagorinsky model

		profs.mix_len_hori_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dxy)));
		profs.mix_len_vert_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dz)));

-        profs.mix_len_hori_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dxy)));
-        profs.mix_len_vert_sq = pow2(real_t(1) / (real_t(1) / pow2(karman_c * max(k, 1) * dz) + real_t(1) / pow2(smg_c * dz)));
+        // combine wall-based and grid-based length scales using a harmonic mean of their squared values
+        real_t wall_mix_len      = karman_c * max(k, 1) * dz;
+        real_t hori_grid_mix_len = smg_c * dxy;
+        real_t vert_grid_mix_len = smg_c * dz;
+        real_t inv_wall_len_sq      = real_t(1) / pow2(wall_mix_len);
+        real_t inv_hori_grid_len_sq = real_t(1) / pow2(hori_grid_mix_len);
+        real_t inv_vert_grid_len_sq = real_t(1) / pow2(vert_grid_mix_len);
+        real_t hori_mix_len = real_t(1) / (inv_wall_len_sq + inv_hori_grid_len_sq);
+        real_t vert_mix_len = real_t(1) / (inv_wall_len_sq + inv_vert_grid_len_sq);
+        profs.mix_len_hori_sq = pow2(hori_mix_len);
+        profs.mix_len_vert_sq = pow2(vert_mix_len);

		this->k_m[0](this->ijk) = where(this->k_m[0](this->ijk) > 0.1 * pow(this->params.di, 2) / this->dt, 0.1 * pow(this->params.di, 2) / this->dt, this->k_m[0](this->ijk));
		this->k_m[0](this->ijk) = where(this->k_m[0](this->ijk) < 1e-6 * pow(this->params.di, 2), 1e-6 * pow(this->params.di, 2), this->k_m[0](this->ijk));

Aniso smg #172

Are you sure you want to change the base?

Aniso smg #172

Uh oh!

Conversation

pdziekan commented Feb 20, 2024

Uh oh!

pdziekan commented Sep 24, 2024

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 19, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant